Tweak rank/relevance algorithm

[pazpar2-moved-to-github.git] / NEWS

Rank algorithm configurable by 'rank' element inside service. So far
only, attribute 'cluster' is recognized. If cluster="yes", multiple
records inside a cluster boosts higher than single records. This
is default behavior and existing behavior. cluster="no" takes the
average score of each record in a cluster.

--- 1.6.17 2012/09/05

Fix bad re-use of connections (connections with changing proxy should not
be reused).

--- 1.6.16 2012/08/22

Fix a bug introduce in 1.6.15 around the position sorting. It resetted the 
resultset and sorting when the sort order is position. However this will be done on
every client poll, which will make pazpar2 continuing reset and fetching. 
It should only be done on FIRST request where the sort order change. 

Fix an issue on suggestion option: Also disable suggestions on empty string. 

Clean up in turbo marc stylesheet. 

Remove the hardcoded size of termlists.

--- 1.6.15 2012/06/27

New facility: ccldirective may be given in service definition. Allows
CCL parsing to be customized a bit, such as defining names of operators
and, or, not.

New facility: raw record by checksum, rather than offset. The record
command optionally takes checksum which identifies certain record from
a database.

New facility: per field ranking. Rank may be given as M [F N] where
M is default rank and N is rank for CCL terms from field F.

--- 1.6.14 2012/06/04

Fix for IE7/8 in pz.js

Applied patch from Giannis Kosmas on keepAlive, which also adds keepAlive to init response. 

Lower log level some places.

Remove some invalid test results. 

--- 1.6.13 2012/05/23

Introducing a version=2 parameter for show, termlist and bytarget commands.
This enables pazpar2 to return approximation on hit and count count when 
doing record filtering using the limit parameter on search and a 
limitmap with a value of "local:"

Setting pz:xslt may embed local XSLT as an alternative to referring
to filenames.
Value is not CDATA but XML nodes embedded, so escaping is not necessary
but a root element *must* be present. For example:
<settings target="target="z3950.indexdata.com/marc">
     <set name="pz:xslt">
       <xsl:stylesheet
          version="1.0"
        ..
       </xsl:stylesheet>
     </set>
    </settings>

Metadata field rank may given by XML internal document (pz:xslt
result). If rank is not given, the rank from service description is
used - as before.

Metadata field can now configured a default limitmap and facetmap. 
Setting limitmap to "local:" would work for all kind of targets, but would
prob. not be the optimal solution. But at least better than the default behavior 
of pazpar2 where no filtering is done. 

A service definition can now also contains <set/> that defines service-wide
settings. These will override server-wide sets and will be overridded by 
settings.

New setting, pz:present_chunk, that specifies number of records to fetch
at a time. Zero will disable chunkation; will fetch max_records at once.

--- 1.6.12 2012/03/14

Revert the format change in termlist response, that could break 
some clients / UIs since they were expecting an (empty) element 
if no facet values was found. 

--- 1.6.11 2012/03/07

Revert the behavior of returning errors when unable to block 
on termlist, bytarget and search, when unable to block due to 
other block. The client will now receive a regular response, 
but it will be logged in the server. A parameter (report) is
added to change behavior to return error response or WARNING 
status message. Consider this "API" as private, as it is mostly 
untested and could be changed in future releases. 

Fix spell error in pz2.js fix in 1.6.10.

New Marc2TurboMarc.xsl (contribution from Sven Porst). 
Can solve the missing marc21.xsl updates in some cases. 

tmarc.xsl: Simplify the 6xx to subject-long and fix 1-based 
substring (contribtion from Sven Porst)

marc21.xsl: fix 1-based substring call

tmarc.xsl and marc21.xsl: use 856$a as last option for electronic-text. 

Add test_termlist_block to test suite

--- 1.6.10 2012/02/23

Fix SEGV for invalid PQFs and SRU/SOLR targets
Also refactor a bit the code that converts from PQF to SRU/SOLR queries.

Fix pz2.js: "null object" due to change in in bytarget result XML.

Fixes in tmarc.xsl: Subject-long shorten for extra commas only. 
Added this normalization to the other subject-long fields (d6xx), 
where it was missing.

Fixes in marc21.xsl: Updated with most of the new tmarc.xsl. 
Still differences around medium and holdings. marc21.xsl is not 
longer active used by Index Data, and should be considered unsupported.
Use tmarc.xsl instead. 

--- 1.6.9 2012/01/19

Fix SEGV that could occur for failed connections.

--- 1.6.8 2012/01/17

Fix bug for command sort that could return no results for active clients
(from previous search). This bug was present in 1.6.6-1.6.7.

Fix bug in results that could include results that should have been
filtered out. This bug was present in 1.6.6-1.6.7.

--- 1.6.7 2011/12/23

Fix bug introduced in 1.6.6 where a connection re-use could stall
incoming requests.

Local filtering may now specify a local metadata field, eg.
 pz:limitmap:somefield[t]=local:otherfield

--- 1.6.6 2011/11/25

For search, when limit and or filtering is in place and search
is identical to previous search, the result set is re-used and the
target is not searched.

Limits may work perform local filtering as well, by using "local:"
for a limitmap.

--- 1.6.5 2011/11/01

Updated bytarget command to contain a suggestions element with misspelled
words and suggestions to these. pz2.js has been updated to deliver this 
onwards as well. Only target that currently delivers this is the solr 
client in YAZ 4.2.18. 

--- 1.6.4 2011/10/28

New service definition element, xslt, that allows an embedded stylesheet
to be defined. This can be referred to from pz:xslt as an alternative to
external files.

New pz:sortmap:field setting for specifying hints on how to make
a target natively sort on a field. This is used for command=show in
conjunction with sort.

New pz:url setting for specifying the actual URL for a target. When
this is used the target ID is not used as URL anymore and the target ID
may be almost any string (not including []).

command=termlist without name parameter returns all termlists/facets.
Previously if name parameter was omitted, only "subject" was returned.

--- 1.6.3 2011/09/22

Make termlist sorting stable. Terms with same frequency are now sorted by
their display name. This makes a pretty display and improves our
regression test because qsort is not a stable sort.

New sort parameter value 'position'. The 'position' sorts merged records
by their original position from the remote target. This is primarily useful
for debugging and may be used for targets that already perform some kind
of relevance ranking. Note that sort by default is decreasing; so to get
records in their original order sort=position:1 must be used.

--- 1.6.2 2011/09/15

tmarc.xsl: yet another 773$g fix. Was broken in 1.6.1 as well.

Facility to change working directory for pazpar2 daemon. Option -wdir
sets working directory to dir. This facility is useful if core dumps
must be saved. In this case, the current working directory must be
writable by the running user, such as "nobody".

--- 1.6.1 2011/09/14

New configuration element <icu_chain> for <server>/<service> which
allows a named ICU rule (chain) to be defined. The names relevance,
sort, mergekey and facet are used for those operations. The definition
  <icu_chain id="sort" locale="en"> ..  </icu_chain>
is identical to
  <sort> <icu_chain locale="en> ... </icu_chain> </sort>
And so on.. for relevance, mergekey and facet as well. The latter
style is deprecated. The facet terms are normalized by the facet
rule by default. This may be changed on a metadata field basis by
defining the new attribute 'facetrule' for the metadata element.
For example:
  <icu_chain id="myrule" locale="en"> ... </icu_chain>
  <metadata name="author" termlist="yes" facetrule="myrule"/>

Preserve rorder for merged metadata. Fixes issue as reported by Sven
Porst: http://lists.indexdata.dk/pipermail/yazlist/2011-July/003230.html

tmarc.xsl: set journal-subpart to 773$ only.

--- 1.6.0 2011/09/12

Modify the behavior for the limit parameter (first defined in 1.5.7).
Mapping of limit searches are now defined by the new configuration item
pz:limitmap. Fix a dead-lock problem with the limit parameter.

Extend tmarc.xsl to extract 773$g data (OpenURL).

--- 1.5.7 2011/08/19

ICU default maps remove backquote (`).

Command 'search' takes limit parameter (optional).  The limit parameter
allows a search to be limited one or more facets and the corresponding
values. This is for server side filtering.

Configure tweak: Use -lm for log(3) if needed

--- 1.5.6 2011/04/15

Fix a problem with skiparticle sortkey that could be completely
ignored (and reduced to "").

Fix dependency problem in pazpar2 RPM package (did not require
libyaz4 as it should).

--- 1.5.5 2011/03/28

Fix memory leak that occurred for command=termlist&name=xtargets .

Pazpar2 may save HTTP requests. Enabled by option -R.

--- 1.5.4 2011/03/08

Experimental support for DTIC DADS target. New dads-pz2.xsl.

Support for query_syntax (overrides the default for SRU | Z39.50) 

Support for extraArgs (ZOOM "extraArgs" option) for targets

New commands: status-server and status-session

--- 1.5.3 2011/02/18

Fix for threaded runs: Client now have a copy of the database URL, 
which can used after the database has been release from the client. 
This makes the logging in the connection idle timeout of the client nicer (no NOURL) and should be thread-safe.  

tmarc.xsl: Add journal-title-abbrev and full text. 

cf.xsl: new fields: isbn, issn, journaltitle, volume, issue

Fix for cmd=record before search.

Session Logging clean up.

Fix wrong termlist factor when maxrecs is different from 100.

--- 1.5.2 2011/01/18

Fix missing pz:termlist_term_factor in settings.c messed up pz:preferred.
Term factor is default enabled but can be diseabled by
pz:termlist_term_factor=0

--- 1.5.1 2011/01/06

Add scaling of facet count. Currently always enabled, needs fixing.
Allow user-defined info for target suffix. This has no meaning in
Pazpar2 except to distinguish targets from each other. The suffix
data begins with #. For example z3950.indexdata.com/gils#Mydata

Added exact-match recordfilter; format name=value

--- 1.5.0 2010/10/05

SOLR support. Pazpar2 may operate as web service client for SOLR.

--- 1.4.4 2010/08/17

Fix for show command and block=1 (dead lock). Bug was introduced in
1.4.0.

--- 1.4.3 2010/07/02

New RPM packages: pazpar2, pazpar2-js, pazpar2-doc. These have been
tested on CentOS 5.5 only.

--- 1.4.2 2010/06/18

Fix problem with result sets being removed from a client session
if the connection for it was resused by another session. Bug #3489.

New iphone UI for Pazpar2 (www/iphone).

--- 1.4.1 2010/05/21

Fixes for threaded operation.

New stylesheets for TurboMARC: tmarc.xsl and opac_turbomarc.xsl.

New example services in etc/services in source. In the Debian packages
these are located in /etc/pazpar2/services-available

Threaded mode operational on Windows. Requires Windows 7 or Windows
Server 2008.

Default value of setting pz:max_connections is 0 which means that there
is no limit on number of connections.

--- 1.4.0 2010/03/29

Pazpar2 may operate in threaded mode. Enabled by element threads in
the configuration. See pazpar2_conf for details.

New setting setting: pz:max_connections. Setting pz:max_connections is
a limit of number of sockets to a host. When this limit is reached,
Pazpar2 will wait up to 5 seconds for a connection to becomes available.
The client will be marked Client_Error when it can not be searched
(other clients in a session may work). If pz:max_connections is not set
for a target, a value of 30 will be used. Note: the pz:max_connections
will only work in threaded mode.

pz2.js: JSON support for show.

Debian package: Enable default service, default.xml, before starting
Pazpar2 only if there is no service already in /etc/pazpar2/services-enabled.

--- 1.3.0 2010/01/25

Debian version depends on on libyaz4. Note that Pazpar2 will still
compile from source with YAZ 3.

Split services into separate files. The example configuration file
pazpar2.cfg.dist now includes a default service default.xml (part of
etc). And default.xml includes settings/edu.xml. The default.xml file,
not to be confused with settings/defaults.xml, is a template for jsdemo
and other services. The Debian package installs /etc/pazpar2/server.xml
which is now the main pazpar2 configuration (used to be called pazpar2.cfg).
server.xml includes services from /etc/pazpar2/services-enabled/*.xml .
The default.xml (from etc) is installed in /etc/pazpar2/services-available
and a symlink to it is created from services-enabled. The default.xml
service is unnamed and, thus, will be used by jsdemo and test1.

New setting pz:negotiation_charset. Patch from Andrei V. Toutoukine. The
new setting pz:negotiation_charset specifies character set for Z39.50 Init.  

--- 1.2.4 2010/01/15

Support for additional fields in cf.xsl and pazpar2.conf.dist:
publisher, available, due, location (=locallocation), callno
(=callnumber), thumburl and score.

Describe pz:xslt and the auto setting.

Move mergekey definition away from the normalization stylesheets and
define a mergekey common for all target types in pazpar2.cfg.

Code update: Use the Odr_int type for hit counts. This is part of
YAZ 3.0.47 and later and so configure checks for that.

--- 1.2.3 2009/11/24

Metadata attribute 'skiparticle' also works for ICU based
normalization. (was only working for the non-ICU/ASCII before).

Command bytarget with argument settings=1 will show settings per
target.. This is to be able to verify correct settings and be able to
test that they are correct. The database settings array size is now
also stored..  Problems with database settings array is that if not
careful it will be too small (smaller than dictionary per-service
in settings).

Make record list sorting stable by comparing mergekey for records if
relevance/title or other sorting criteria all match. This is merely to
ensure that our regressions tests works (reproducible output).

Relevance calculation changes: use a different denominator (length) for
per-field relevance scoring.. Instead of length of all ranked fields we
now use length of individual fields (as if they were individual "free"
text fields). This will ensure that documents with a long field with no
match (say description) will not "hurt" a title match.

Diagnostic member was not set on connection error. Fixed

--- 1.2.2 2009/10/13

Command search takes two optional parameters, startecs and maxrecs,
that specifies the start offset (0, 1, ...) and maximum number of records
to fetch for each target.

XSLTs + MARC maps are cached within a session so we don't re-parse
them over and over again. Even for a session with a single search
there's much to be gained because many targets use the same
transformation.

The metadata attribute 'mergekey' now takes one of three values 'no',
'required', 'optional' .  And the resulting mergekey from metadata
is now ordered in the same way as metadata in the service definition.
Older Pazpar2 version use the order in which metadata appeared in a
record instance.

The search argument 'filter' now offers a new operator ~ which does a 
substring match. The = operator works as before: string match for
anything but pz:id, or target match for pz:id.

New setting pz:recordfilter. The value of this setting takes the
form name[~value]. This setting makes Pazpar2 ignore all retrieved
records that do not have the metadata element name with value substring
(if given).

Pazpar2 allows YAZ log level to be set (option -v).

--- 1.2.1 2009/10/05

For WS responses Pazpar2 creates XML header. Exception: raw record.

Setting XML files are now stored in etc/settings instead of etc. This
reflects the layout with the Debian package layout.

Settings may be posted for command=settings. The POSTed settings must
have root element 'settings' like regular setting files. In order to be
recognized, the POST request must use Content-Type=text/xml.

A service may be posted for command=init. This service will be used
during the session. The service may have its own target settings,
ICU config, timeout, etc. In order to be recognized, the POST request
must use Content-Type=text/xml.

Timeout values may be given per-service. That's element 'timeout' 
which takes three attribute values (a subset may be given): 'session',
'z3950_operation', 'z3950_session'. Option -T is no longer supported
- used to specify session timeout.

Option -t tests the Pazpar2 configuration and returns exit code
(0=success, non-zero=failure). In previous version of Pazpar2, -t
specified local settings.

In version 1.2.0 the configuration file - after include processing -
was dumped to stdout. Now, the configuration is only dumped to the
yaz log file if option -d is given.

--- 1.2.0 2009/09/10

Configuration may now have multiple server areas. This means that a
Pazpar2 instance may listen on multiple ports. Virtual hosting is not
yet supported - on a server basis. Configuration may also have multiple
services .. That is repeating service elements inside a server. Each
has an attribute 'id' which serves as service ID. This ID in turn may
be used in a Pazpar2 session, by specifying parameter service=ID for
command init. There can be at most one unnamed service inside a server
which can be referred to by not specifying an service ID for command
init (backwards compatible). In order to partition multiple servers and
services a new include directive has been added. This takes an attribute
'src' which specifies one or more sub-files. For example to include
service files, one might use:
  <server >.. <include src=/"etc/pazpar2/conf.d/*.xml"/> .. </server>.
It is the intention that that completely makes the settings directive
redundant.

Fix problem where the record command would wait forever if there were
no targets to wait for (activeclients == 0).

--- 1.1.1 2009/08/28

One result set is created per session (last search) rather than for
each connection which happen to be shared (bug #3009).

marc21 stylesheets changed for efficiency.

--- 1.1.0 2009/08/19

Session timeout may be specified on the command-line as option -T.

Pazpar2 may now be operated in a no-merged mode for records.. All records
will be considered unique. This mode is enabled if no mergekey is
generated by the normalization stylesheet (pz:xslt). 

Pazpar2 caches original records from each target and the 'record' command
with offset returns the original record if 'syntax' and 'esn' are NOT
specified. This speeds up retrieval of original records but also means
that Pazpar2 uses more memory. The cached records will be freed when the
session terminates or a new search is executed.

Pazpar2 no longer uses its own ICU wrapper. Instead the ICU wrapper
library part of YAZ is used.

Added SRU client support.

Automatically computes pz:nativesyntax if not provided. Works for XML and
MARC.

--- 1.0.13 2008/11/24

Command bytarget returns name of target (if defined).

--- 1.0.12 2008/11/04

Fixed bug #2021.. location now holds all brief elements.

--- 1.0.11 2008/10/15

Fixed check for application/x-www-form-urlencoded parameters.

--- 1.0.10 2008/10/14

Fixes for IE in pz2.js.

Fixed bug #2021: non-merged, brief meta data NOT included for command=show.

--- 1.0.9 2008/05/23

Changed the JS library pz2.js to use POST for long URL (+ params).

Added installation instructions for Windows. Note: NT services is
NOT available until we make a new release of YAZ.

Preserve order of repeated metadata fields (they were reversed before).

More MARC21 information extracted for metadata.

--- 1.0.8 2008/04/22

Fixed bug #1162: HTML entities are not escaped properly.

Native Windows port of Pazpar2. Makefile for Visual Studio provided.

--- 1.0.7 2008/04/14

Marc21 stylesheet updated to reflect multiple full text fields

--- 1.0.6 2008/03/12

Fixed bug in pz2.js WRT DOMElement attributes on IE.

Fixed bug 2100: Database wildcards not working

--- 1.0.5 2007/11/09

Added support for retrieval of records in binary.

Fixed bug 1794: Pazpar2 does not return valid XML.

Deal with ICU not returning sortkey (resulted in SEGV before).

--- 1.0.4 2007/09/28

JavaScript library pzw2.js throws error if WS response (from Pazpar2 or
other) is malformed (non-wellformed XML or missing Pazpar2 OK status).

Improved diagnostics when Pazpar2 HTTP decoding fails.

Pazpar2 requests may be POSTed as using Content-Type
application/x-www-form-urlencoded.

Pazpar2 honors LF in HTTP headers.

Handle targets that handle negative hit counts (should not happen, but it
does).

--- 1.0.3 2007/09/19

ICU is used for tokenization and normalization of the following: mergekey,
sorting, relevance terms.

Debian package now enables ICU tokenization and normalization by default.

--- 1.0.2 2007/08/22

Exposed user setting values (i.e. non-pz: names) to the record systems in two
ways: Either as parameters to the normalization stylesheets (which would allow the
programmer to postprocess or use the values in any way) or after the normalization
step, in which case values are made part of the normalized record (and available for
sorting, termlists, display, or other interface-related use.

Implemented sorting by year.

Option -d dumps records to the current log file instead of stderr.

Fixes for compilation on cygwin.

Z39.50 client code uses pz:elements. pz:elements was recognized in
earlier Pazpar2 versions but it was not used for anything.

icu_chain_test is using fgets instead of getline - fixes compilation
on MacOSX .

Loosen the CCL query parsing so that Pazpar2 only returns error if _all_
query conversions fail (rather than _any_). This means targets that do
not support some fields are ignored in a search.

--- 1.0.1 2007/07/03

Improved handling of socket timeout for Z39.50 connections. 

Misc documentation updates and spell fixes.

Debian package pazpar2 creates log rotate entry.

Debian package pazpar2-apache2 reloads Apache2.

jsdemo included in distribution. It illustrates the use of the js/pz2.js
JavaScript library.

--- 1.0.0 2007/06/20

First public release.